What is submetering?

The mechanics

The default condition in most facilities is that electricity is measured once, at the service entrance, and everything downstream of that point is allocated by assumption — by floor area, by headcount, by nameplate rating, or by whatever was in the lease when it was signed. Submetering replaces the assumption with a measurement.

A submeter is a current and, where it matters, voltage measurement at a specific point in the distribution system. Typical measurement points, in descending order of how much they usually justify:

• Billing boundaries — the demarcation between a landlord's supply and a tenant's premises
• Major loads — chillers, air handlers, compressors, the top 20 or so consumers by kW
• Production lines or cost centres — where cost has to be assigned to output
• Panels and feeders — where the objective is to bound a system rather than identify one asset

The two grades, and why blurring them is expensive

There are two entirely different accuracy requirements in this field, and most of the trouble in submetering projects comes from treating them as one.

	Billing grade	Insight grade
Where	Billing boundaries only	Everywhere downstream
Equipment	Externally powered meters with their own voltage reference and metering-class CTs	Self-powered wireless current sensors
Accuracy target	ANSI C12.20 accuracy classes; GSA's billing threshold is ±2%, revenue grade ±0.5%	GSA's fault-detection threshold, ±10%
Legal	Jurisdiction-mapped before hardware is quoted	Explicitly non-billing, stated in the contract

The reason this matters is measured, not asserted. GSA field-tested the self-powered wireless CT sensor class and published 7% average energy measurement error, with individual readings ranging from 52% under to 38% over depending on load type, an inability to measure below 0.75–1 amp, and no power factor without a paired voltage reference.

That performance is entirely adequate for fault detection, where GSA's own threshold is ±10%, and entirely inadequate for billing, where its threshold is ±2%. Both statements are true at once. A vendor that puts one sensor class on both jobs is either overspending on insight points or under-delivering on billing points, and the second one surfaces in a chargeback dispute rather than at install.

(GSA's separate Single-Circuit Meter Findings reports "<2% measurement error under most circumstances" for a different device class. The two documents are not interchangeable and must not be quoted together.)

Worked example: what it costs and what it returns

GSA's published testbed prices give a defensible cost basis: $141 per current transformer plus a $259 bridge, against $901 per measured load for the comparator technology at the same site. Installation labour was documented at roughly 8 hours for 144 CTs across 13 panels and 4 HVAC disconnects.

For a 40-space industrial park at billing-grade pricing of $1,500–$2,000 per submeter:

• Hardware: 40 × $1,750 = $70,000
• Suppose the landlord is currently absorbing $50,000 a year of utility cost that a measured allocation would recover
• Recovered expense flows dollar-for-dollar to net operating income, and net operating income capitalises: at a 7% cap rate, $50,000 = $714,000 of asset value; at 6%, $833,000

The only inputs to that arithmetic are the owner's own recovery delta and the owner's own cap rate, which is why it survives diligence. Be direct about the limit of the claim: there is no strong independent ROI study for commercial submetering. The arithmetic is the case, not a borrowed statistic.

The standards argument nobody makes

Whole-facility measurement and verification — IPMVP Option C — cannot reliably detect savings below roughly 10%, because the signal is buried in the building's own variability. A project that genuinely saves 6% is unprovable at the whole-building meter.

Submetering the affected system converts that into an Option B problem: isolate the system, measure it before and after, and the signal-to-noise ratio improves by an order of magnitude. That is the standards-grounded reason submetering exists, and it directly supports 179D's measurement pathway and the M&V obligation in every energy savings performance contract.

The five things that actually go wrong

None of them is sensor accuracy.

1. Unmapped circuits. GSA's own wording: identifying which loads are associated with which circuits is difficult "because of inaccurate panel schedules, obscure naming conventions, or lack of circuit tracing." Everything downstream inherits the error, and bad point mapping is the most common cause of a credibility collapse in month three.
2. Wrong CT sizing. Size to estimated load, not breaker rating. The measured load should be at least 10% of the CT's rating; a CT sized for a 400 A breaker on a 12 A circuit reports noise.
3. Phase-angle error at low power factor. At 0.45 power factor, a 3.3° CT phase error produces a 13.1% power error — worst on exactly the lightly-loaded motors most worth investigating.
4. Wrong phase mapping or polarity. Reversed polarity, wrong vector sequence and voltage/current phase mismatch all produce confidently wrong numbers rather than obvious failures.
5. Clock skew. Beyond 3 minutes the data needs proration; beyond 75 minutes it should be estimated rather than used.

What submetering is not

It is not tenant billing. Submetering is measurement. Tenant billing is a legal and commercial process built on top of it, governed by state utility law and by the lease. A submeter is necessary and nowhere near sufficient. See what is tenant billing.

It is not required for ENERGY STAR or for benchmarking filings. Those need twelve consecutive months of whole-building consumption for every fuel, and monthly bills satisfy that. Any vendor claiming submetering is required for a Portfolio Manager score is misrepresenting the requirement. The genuine gap in multi-tenant property is tenant-metered load, which is a real submetering case — but it is a different argument.

"Revenue grade" is not a property of a sensor. Certification covers the entire measurement chain, including the current transformers, and it is jurisdiction-specific. A claim of revenue grade that does not name the CT and the type approval is unsupported.

More points is not better. The top 20 loads by kilowatt plus the tenant boundaries capture most of the available value. Metering every branch circuit multiplies cost and mapping burden, and the self-powered sensors cannot read below roughly 0.75–1 amp anyway.